On 10/26/22 01:42, Alex Bennée wrote:
Not a problem with this patch but I was a little confused by PAGE_VALID
because its the one "special" flag not documented in cpu-all.h:
/* same as PROT_xxx */
#define PAGE_READ 0x0001
#define PAGE_WRITE 0x0002
#define PAGE_EXEC 0x0004
#define PAGE_BITS (PAGE_READ | PAGE_WRITE | PAGE_EXEC)
The above are self explanatory as they mirror the mmap flags. But what
does PAGE_VALID really mean.
It exists so that unmapped pages have flags == 0 and
mmap(..., PROT_NONE, ...) pages have flags != 0.
Perhaps a better name would have been "allocated".
You are perhaps 25 years to late to bikeshed that name. :-)
The only place
that ever seems to clear the flag is the PPC mmu_helper code in a
response to a particular TLB operations. Should that code instead be
doing page_set_flags(PAGE_RESET)?
Heh. ppc seems to have re-used the symbol for its own internal data structure.
Well, considering that it's got to pass on "prot" to tlb_set_page at some point, re-using
those names isn't a horrible idea.
Anyway, you can see target_munmap clear PAGE_VALID here:
page_set_flags(start, start + len, 0);
PAGE_RESET exists to distinguish mmap replacing an exiting mapping (old backing storage
replaced) from mprotect (old backing storage retained, new permissions applied).
r~
